In a sit-down with Acuity, Camber’s CEO said many provider AI investments may have lost money this year, and explained why LLMs stay out of claim submission.
Key Takeaways
- Apollo’s skepticism matches what Camber sees: Rimann told Acuity that early 2026 AI spending by larger provider organizations may, in many cases, have produced a negative return.
- Rebuild the process, not the bottleneck: Automating one bottleneck only moves it elsewhere, he said, so Camber redesigns the workflow before deciding where AI belongs.
- Claims go out deterministically: Camber uses LLMs to research how claims should be built, then submits them through deterministic pathways, with people reviewing the AI’s findings.
- Client-certified results: Rimann said the 20% to 50% EBITDA figure is a platform average, measured mainly by collections uplift that customers’ owners or CFOs certify.
Christophe Rimann does not dispute Apollo’s numbers. When Torsten Slok, the Chief Economist at Apollo Global Management, published an Oct. 24 note titled “No Signs of AI Payoff Outside the Tech Sector,” Rimann, Co-Founder and Chief Executive Officer of the revenue cycle company Camber, wrote on LinkedIn that the finding was “honestly completely in line with what we see on the front lines of AI.” In the same post, he described a different result for Camber’s own customers: EBITDA increases of 20% to 50%.
In an interview with Acuity, Rimann walked through both halves of that post: why he thinks so many healthcare organizations are not yet getting a return on AI, and how Camber measures the gains it says its customers are seeing.
“AI is an incredible tool. It’s incredibly powerful,” he said. The difficulty, in his telling, lies less in the technology than in what organizations build around it.
Why Camber’s CEO Agrees With Apollo on AI’s Missing Payoff
Slok, who is also a Partner at Apollo, built his case on a split between the seven largest technology companies and the rest of the S&P 500. Revenue per worker and profit margins have risen for the former but stayed flat for the remaining 493 companies, according to his charts, and a utilization-adjusted measure of total factor productivity maintained by the Federal Reserve Bank of San Francisco shows no sign yet of AI lifting productivity growth. Mainstream companies, he wrote, are starting to treat spending on AI model usage, measured in tokens, as a cost center rather than a source of growth. “AI may be a great calculator, but in business, intelligence is mostly judgment,” he wrote.
Rimann described the same pattern from inside healthcare. From early this year into roughly May or June, he said, many companies focused on maximizing token usage, and spending rose sharply. “In many cases, some of the AI investments for larger provider organizations might have ended up negative in the ROI sense,” he said. Many provider organizations, in his view, were not sure how to get value out of the tools they were paying for.
He pointed to MIT research from last year as reaching a similar conclusion. That report, “The GenAI Divide: State of AI in Business 2025” from MIT’s Project NANDA, found that 95% of organizations in its research were getting no measurable profit-and-loss return from generative AI pilots, based on interviews with 52 organizations, a survey of 153 senior leaders, and a review of more than 300 public AI initiatives.
Part of the problem, Rimann said, is that early impressions of the technology are misleading. “You hop on Claude or you hop on ChatGPT, and immediately you’re starting to see personal value,” he said. As use spreads across an organization, coordination costs grow, he said, and things start to break down at scale.
Camber’s Background in ABA Revenue Cycle Management
Camber’s business sits in one of the most rule-bound corners of healthcare: getting paid by insurers. The New York company, which operated as Juniper Behavioral Health before taking its current name, was part of Y Combinator’s Winter 2021 batch and built its platform first for applied behavior analysis providers. In February 2025, it announced a $30 million Series B led by Andreessen Horowitz, with participation from ACME, Craft, K Ventures, and Y Combinator, bringing its total funding to $50 million.
A vendor listing Camber maintains with the Association for Community Affiliated Plans says its platform supports more than 250,000 patients and processes more than $2.5 billion in claims annually across 42 states, which the company describes as roughly 20% of the U.S. ABA market. Acuity profiled Camber’s predictive billing model in March, and in September the company acquired Async Lambda, an ABA workflow automation firm.
Why Automating AI Bottlenecks Turns Into Whack-a-Mole in Healthcare Claims
The most common use of AI today, Rimann said, is to get rid of whatever task is in front of someone, which is how most technologies are first adopted. That approach can work well in fields such as law, he said, where a single document can be handed to a model. Healthcare is different because of its context and workflow complexity, he said, which makes it hard to hand off one component in isolation. “Working claims is tough, but understanding the whole context of a claim is even tougher,” he said.
Camber made the same mistake early on, Rimann said. Three or four years ago, when the company began using AI aggressively, it tested the technology against individual bottlenecks. “The realization is rebuilding the process to be AI native is what matters a lot,” he said. Automating bottlenecks one at a time “ends up as a game of whack-a-mole,” he said, because each fix pushes the bottleneck somewhere else in the process.
His LinkedIn post made the same argument through history, comparing early AI adoption to factories that swapped steam engines for electric motors without changing anything else. The comparison draws on the Stanford economic historian Paul A. David, whose 1990 paper “The Dynamo and the Computer” traced how U.S. manufacturing productivity surged only in the 1920s, decades after electrification began, once factories were redesigned around individual motors and the flow of materials.
Rimann’s method for rebuilding starts with the process rather than the model. “We actually try to remain very skeptical that AI is the right tool to begin with,” he said. Camber first works out how a process should run, he said, and only then decides whether and where AI belongs in it. That ordering echoes advice other behavioral health technology executives gave Acuity earlier this year: that automating a broken process only speeds up bad outcomes.
Inside Camber’s Revenue Cycle: LLM Research and Deterministic Claim Submission
Camber uses AI across its operations, Rimann said, including research into how to submit better claims, pulling electronic remittance advice from payer portals, and conducting phone calls with payers. The step most people would assume is automated with AI is the one that is not. “How we submit claims is extremely deterministic. We don’t leverage LLMs for that,” he said.
The division is deliberate. “We do a lot of LLM-based research to understand how to submit claims better,” Rimann said, but the submission itself runs on fixed, rule-based pathways because “that allows us to make sure we’re driving the outcomes we want very repeatably.” Camber did try using AI to speed up claim submission, he said, and found it did not improve results. Deterministic submission, in his account, is faster, cheaper, and better for customers.
Human Review in an AI-Native Revenue Cycle
Rimann rejected the idea that the goal is to remove people from the work. “It’s not how do we get humans out of the loop entirely. It’s how do we rethink what humans are responsible for in a day and age where AI can do so much and so powerfully,” he said.
In the older model, people submitted claims and reviewed denials. In Camber’s version, deterministic pathways handle submission, while AI assembles context so that a person can decide how claims should go out at scale. The aim, he said, is to build that into the system so that the next 100,000 claims carry “AI-generated, human-reviewed insight.”
He compared the shift to agentic coding, in which developers hand most of the writing to AI tools and change what they spend their own time on, a change that has also reshaped the build-versus-buy math for ABA providers. “Humans are much more around architecture, much more around trying to do the creative problem solving,” he said, with review of the output as the final responsibility. In revenue cycle work, that review is meant to ensure claims go out correctly and in a compliant way.
Paul McDonnell, Head of Marketing at Camber, joined the conversation, and pointed to a June essay by Microsoft Chairman and Chief Executive Officer Satya Nadella on the future of the firm, which argues that a learning loop between people and AI becomes “the new IP of the firm.” McDonnell described Camber’s approach as pragmatic: identifying the core business opportunity first, then deciding where AI applies.
How Rimann Says Camber Measures EBITDA Gains for Providers
Rimann addressed the figures in his post directly. The 20% to 50% EBITDA increase, he said, reflects “the average numbers across the platform,” which he put at more than 100,000 providers across the United States in 10 to 12 specialty verticals. Camber started with a narrow focus, he said, and over the past couple of years has expanded into new verticals while driving similar results.
Those results are confirmed by customers themselves, according to Rimann. “It’s actually either the owner-operator or the CFO who’s kind of certifying these results,” he said, signing off on the value attributable to Camber. The main yardstick is the uplift in a client’s collections rate. “How much more money are we helping you collect from insurance payers this year?” he said.
In his post, Rimann wrote that on a run-rate basis, those gains will exceed $100 million across Camber’s customers over the next year, and that the money is going directly into expanding access to and affordability of care across the country.
That yardstick maps onto what ABA owners and their investors watch most closely. Mergium Advisors counted roughly 93 private equity-backed platforms in pediatric therapy at midyear, as Acuity reported in July, and collections flow directly into the EBITDA those owners underwrite.
Providers say they want that kind of evidence. In Experian Health’s 2026 State of Claims survey of 204 healthcare professionals responsible for financial, billing, or claims decisions, 75% said denials were increasing, 40% said more than 10% of their claims were denied, and 83% said they would replace existing technology if a compelling return on investment were demonstrated. Experian Health, which sells AI-based claims tools, also found 64% planned to invest in AI for claims and denial management. Mindy Fortson, its Chief Operating Officer, said in the release that providers’ “response isn’t keeping pace with the problem,” citing concerns about cost, return on investment, and the speed of AI’s evolution. The full report is available with registration.
Vendor durability belongs in the same calculation. ABA providers that had built billing workflows on one automation vendor were given roughly 90 days to replace it when Thoughtful AI wound down its stand-alone services this spring.
Camber’s Next Question: Plugging Holes or Redesigning the Ship
Near the end of the conversation, when Acuity compared organizations patching their operations with AI to a crew plugging holes in a sinking ship, Rimann said he expected to publish an internal memo at Camber within days on nearly that question. “Are we plugging holes in a ship, or are we trying to think through how the ship should operate?” he said.
Slok’s note argues that the AI story will be decided in the S&P 493, by companies that have yet to show the technology in their margins. Rimann’s answer is that the gains come to organizations willing to redesign the work first. Whether Camber’s customers bear that out is a question their collections, more than any LinkedIn post or five-chart note, will eventually settle.
